Visual Understanding under Complex Environments and Limited Data
复杂环境与数据受限下的视觉理解
| Chair: |
Co-chair: |
 |
 |
| Xiaoxu Li |
Minghua Zhao |
| Lanzhou University of Technology, China |
Xi’an University of Technology, China |
| |
|
| Summary: |
|
- Visual Understanding under Complex Environments and Limited Data is a highly challenging research direction in computer vision. Real-world applications often face a dual dilemma: on one hand, complex and changing environmental factors such as drastic illumination variations, adverse weather (rain, snow, fog), dynamic backgrounds, occlusion, and target deformation severely disrupt the stability of visual perception. On the other hand, high-quality annotated data is severely limited—for example, in specialized fields like medical imaging, remote sensing analysis, and industrial inspection, obtaining large-scale precise annotations is costly or even infeasible. This forum aims to explore how to enhance the robustness and generalization ability of visual models under harsh environments and scarce annotation conditions, pushing visual understanding technologies toward more realistic and deployable applications. The forum will focus on, but is not limited to, the following directions: self-supervised and semi-supervised learning, domain adaptation and generalization, generative data augmentation, meta-learning and few-shot learning, multimodal joint training, and uncertainty modeling.
|
| |
|
- 复杂环境与数据受限下的视觉理解是计算机视觉领域极具挑战性的研究方向。真实应用场景往往同时面临两重困境：一方面，光照剧烈变化、恶劣天气（雨、雪、雾）、动态背景、遮挡与目标形变等复杂多变的环境因素，严重干扰视觉感知的稳定性；另一方面，高质量标注数据严重受限，例如在医疗影像、遥感分析、工业检测等专业领域，获取大规模精准标注成本高昂，甚至无法获取。本专题论坛旨在探讨如何在恶劣环境、稀缺标注条件下提升视觉模型的鲁棒性与泛化能力，推动视觉理解技术走向更真实、更落地的应用场景。论坛将关注但不限于以下方向：自监督与半监督学习、域适应与泛化、生成式数据增强、元学习与小样本学习、多模态联合训练、不确定性建模。
|